Improving Heart Disease Diagnosis via a Voting Ensemble Approach

Early detection of heart disease is vital for reducing mortality and improving patient outcomes. This paper proposes a heart disease prediction model based on a Voting Classifier that integrates six machine learning algorithms—GBDT, Random Forest, Decision Tree, SVM, XGBoost, and LightGBM—using a soft voting strategy with weighted emphasis on high-performing models. The dataset, compiled from four medical sources, was preprocessed with Iterative Imputer and mode imputation to address missing values. Experimental results demonstrate that the ensemble model consistently outperforms most individual classifiers across multiple metrics, confirming its robustness and effectiveness. The findings underscore the value of ensemble learning in handling complex medical data and its potential to support intelligent healthcare decision-making.